AI Agents in Engineering Design: A Multi-Agent Framework for Aesthetic and Aerodynamic Car Design

TL;DR

<3-5 sentence high-level summary>This work addresses the challenge of accelerating automotive design by integrating AI Design Agents into a unified multi-agent framework that links conceptual sketches to production-ready 3D geometries, CFD meshing, and real-time aerodynamic evaluation. It combines vision-language models, large language models, and geometric deep learning within an AutoGen-driven architecture to coordinate four specialized agents (Styling, CAD, Meshing, Simulation) and utilize a large-scale DrivAerNet++ dataset of 8,000 designs. A DeepSDF-based CAD representation enables retrieval and interpolation in 3D latent space, while surrogate models enable real-time aerodynamic predictions, drastically reducing iteration times from weeks to minutes. The framework demonstrates practical impact by enabling rapid, data-driven exploration of aesthetics and aerodynamics, with broad implications for engineering design beyond automotive domains.</p>

Abstract

We introduce the concept of "Design Agents" for engineering applications, particularly focusing on the automotive design process, while emphasizing that our approach can be readily extended to other engineering and design domains. Our framework integrates AI-driven design agents into the traditional engineering workflow, demonstrating how these specialized computational agents interact seamlessly with engineers and designers to augment creativity, enhance efficiency, and significantly accelerate the overall design cycle. By automating and streamlining tasks traditionally performed manually, such as conceptual sketching, styling enhancements, 3D shape retrieval and generative modeling, computational fluid dynamics (CFD) meshing, and aerodynamic simulations, our approach reduces certain aspects of the conventional workflow from weeks and days down to minutes. These agents leverage state-of-the-art vision-language models (VLMs), large language models (LLMs), and geometric deep learning techniques, providing rapid iteration and comprehensive design exploration capabilities. We ground our methodology in industry-standard benchmarks, encompassing a wide variety of conventional automotive designs, and utilize high-fidelity aerodynamic simulations to ensure practical and applicable outcomes. Furthermore, we present design agents that can swiftly and accurately predict simulation outcomes, empowering engineers and designers to engage in more informed design optimization and exploration. This research underscores the transformative potential of integrating advanced generative AI techniques into complex engineering tasks, paving the way for broader adoption and innovation across multiple engineering disciplines.

Figures (13)

• Figure 1: AI Design Agents for Accelerating the Car Design Process. The framework integrates vision-language models (SDXL, ControlNet, DALL·E), geometric deep learning models (DeepSDF, PointNet, RegDGCNN, TripNet), and LLMs to automate design tasks. AutoGen helps different agents communicate with each other, improving coordination and making the design process faster and more efficient. The agents can also interact with various engineering tools and execute Python commands, enabling automation of complex design and simulation workflows.
• Figure 2: Our multi-agent system enables effective interaction between engineers and designers throughout the car design process. Given an input 2D sketch for conceptual design, the Styling Agent renders high-resolution images, enhancing visual aesthetics. The CAD Agent generates new designs via generative modeling or retrieves 3D meshes from the DrivAerNet++ database. The Meshing Agent creates high-quality computational meshes for CFD simulations and evaluates mesh quality. Finally, the Simulation Agent performs real-time aerodynamic performance predictions or retrieves aerodynamic data from the DrivAerNet++ database, accelerating the iterative design process. Different outputs from the various agents can be used for design exploration or design optimization, enabling a data-driven and efficient approach to automotive design.
• Figure 3: Example use case of the AI-driven multi-agent system for car design. Designers start with a 2D sketch, which the Styling Agent uses to generate high-resolution renderings. The CAD Agent then retrieves similar 3D designs from DrivAerNet++ for further exploration. Engineers can request aerodynamic analysis, and the Simulation Agent provides CFD simulation data, allowing for rapid performance evaluation and iterative design improvements.
• Figure 4: Continuation of the AI-driven multi-agent system use case for car design. After retrieving similar designs, the CAD Agent generates intermediate shapes through latent space interpolation. These interpolated designs transition smoothly between the estateback and notchback configurations while maintaining geometric consistency. Since these new shapes are not part of the existing dataset, the Meshing Agent generates high-quality computational meshes for aerodynamic analysis. Engineers can then run CFD simulations to evaluate performance, enabling iterative refinement and optimization.
• Figure 5: Different data representations and modalities from DrivAerNet++, a dataset comprising 8,000 industry-standard car designs. These modalities—including 3D CAD models, 3D meshes, point clouds, voxel grids, depth maps, and part annotations—are leveraged by various generative AI models depending on the design task, such as retrieval, 3D reconstruction, styling, and aerodynamic simulation. In this work, we further extend DrivAerNet++ by adding Signed Distance Function (SDF) representations, multi-view images, and sketches to support diverse generative design tasks.